Research Of Audio Signal Recognition System Based On DM6446

Audio recognition is an interdisciplinary subject and has a very wide range,it is the most natural and concise way of communication for man-machine.People usually use Dynamic Time Warping (DTW), Hidden Markov Model(HMM), Vector Quantization (VQ), Artificial Neural Network (ANN) and so on.This paper researched an audio recognition system that based on DM6446.In this article, firstly, it analyzed the common audio processing technology andbasic principle of audio recognition technology, and then it analyzed thealgorithm of audio recognition in three of the most popular ones, includingHidden Markov Model (HMM), Artificial Neural Network (ANN) and VectorQuantization (VQ). The three methods are all have their own characteristics, inview of this paper, it is a small vocabulary audio recognition system, so aftercomparison, this system used the method of Vector Quantization (VQ) as patternmatching algorithm. Vector Quantization (VQ) is the data compression andcoding method that widely used in audio recognition, when realizing the isolatedwords recognition system, it achieved the real-time identification of voicecommands simply, quickly and effectively. Through the system simulation inMATLAB, it got a high recognition rate and explained the reliability of thissystem.In order to detect the real time of the system, the last chapter researched thathow to transplant the Vector Quantization (VQ) algorithm into DSP hardwareplatform to realize the audio recognition, it analyzed the internal structure of theDM6446, the peripheral system, acquisition circuit of the phonetic system anddevelopment environment. The development environment is CCS3.3, whenwriting C program or assembly program in it will get a practical system. After thereal-time experiment, it showed that the system satisfied the real-time require ment and had cert ai n effecti veness.